Geology and geomorphology

The fragmentation of the supercontinent Gondwana commenced in the Early Cretaceous (ca. 136 Ma) as South America was sheared westward along the Falkland Agulhas Fracture Zone (Martin and Hartnady, Reference Martin and Hartnady1986; Eagles, Reference Eagles2007), creating the steep and deep shelf break on South Africa's eastern seaboard that is now only ~8 km wide. The unique morphology of the continental shelf plays an important role in Pondoland's climate and ecology, primarily by preventing the warm-water Agulhas Current from drifting too far from the coastline during the Pleistocene and Holocene (Winter and Martin, Reference Winter and Martin1990; Peeters et al., Reference Peeters, Acheson, Brummer, de Ruijter, Schneider, Ganssen, Ufkes and Kroon2004). The effect of the warm Agulhas Current today drives the convection of moist coastal air cells for coastal precipitation, delivering a source of fresh water to the region (Jury et al., Reference Jury, Valentine and Lutjeharms1993; Schumann et al., Reference Schumann, Cohen and Jury1995). It also influences marine habitats by stabilizing inshore water temperature, which supports intertidal shellfish and littoral fish communities. Paleoceanographic records show that the Agulhas Current may have been up to 4°C cooler during the LGM (Simon et al., Reference Simon, Arthur, Hall, Peeters, Loveday, Barker, Ziegler and Zahn2013; Simon et al., Reference Simon, Gong, Hall, Ziegler, Barker, Knorr, van der Meer, Kasper and Schouten2015a; Simon et al., Reference Simon, Ziegler, Bosmans, Barker, Reason and Hall2015b), but with little variability in the location or strength of the current (Purcell, Reference Purcell2014). These relatively cooler, but sustained, waters offshore Pondoland during the LGM would have continued to maintain suitable habitats for local shellfish communities, particularly those species that were harvested most frequently.

On land, the Pondoland coastline is dominated by various formations of sandstones, shales, and dolerites to the south with granites and gneisses to the north (Fig. 2). Most significant on this coast is a trapezoidal exposure of Late Devonian (385–359 Ma) Msikaba Formation sandstones (Kingsley and Marshall, Reference Kingsley and Marshall2009) that is ~70 km long × ~20 km wide. Inland lithologies beyond the coastal plain are made up of deposits of the Carboniferous to Triassic Karoo Supergroup, including bands of glacial tillites and diamictites of the Dwyka Group, Ecca Group shales, and mudstone and sandstones of the Beaufort Group. Drakensberg Group flood basalts cap the Great Escarpment, which lies ~150 km inland from the Pondoland Coast. Dolerite dykes and sills had penetrated the clastic deposits of the Karoo Supergroup at the time of extrusion of the Drakensberg Group lavas (Duncan et al., Reference Duncan, Hooper, Rehacek, Marsh and Duncan1997).

Figure 2. (color online) Low-altitude panorama of Waterfall Bluff looking westward. Taken with a DJI Mavic Pro, this image shows Waterfall Bluff, the rock shelter, and the surrounding cliffs and landscape. From left to right, the image pans from south to north, respectively.

The Msikaba sandstones are an important feature of this area because these rocks have supported the formation of coastal and riverine rock shelters, which were occupied by hunter-gatherers, and now protect fragile archaeological records, including shellfish and fish remains, that would be more susceptible to deterioration in open-air contexts. This, and the narrowness of the shelf, means that any archaeological site found on or near the modern coastline at Waterfall Bluff would have remained within daily or monthly coastal foraging ranges (5–10 km) throughout glacial maxima (Fisher et al., Reference Fisher, Bar-Matthews, Jerardino and Marean2010). Pondoland, therefore, is one of the few known places across the entire South African seaboard where one can justifiably expect to find continuous records of coastal foraging and fishing that have not, or only to a lesser degree, been influenced by coastline changes.

Pondoland climate and vegetation

Pondoland is part of the Indian Ocean Coastal Belt (IOCB) biome, a well-known center of biodiversity and plant endemism (Van Wyk, Reference Van Wyk, van der Maesen and van Medenbach1996; Mucina et al., Reference Mucina, Scott-Shaw, Rutherford, Camp, Matthews, Powrie, Hoare, Mucina and Rutherford2006). Contemporary rainfall in the region is predominantly during austral summer with precipitation reaching 1,120 mm/yr (Mucina et al., Reference Mucina, Scott-Shaw, Rutherford, Camp, Matthews, Powrie, Hoare, Mucina and Rutherford2006). Under summer rainfall conditions, the amount and character of rainfall in the IOCB is topographically dependent. The northern regions close to the coast have year-round rainfall, while rainfall seasonality increases inland. Today, our study area has one of the highest proportions of winter-rainfall among the coastal areas of the IOCB with up to 36.6% winter rainfall (Cawe, Reference Cawe1994; Mucina et al., Reference Mucina, Scott-Shaw, Rutherford, Camp, Matthews, Powrie, Hoare, Mucina and Rutherford2006).

The Pondoland landscape is characterized by elevated plateaus and deep gorges. The coarse-grained Msikaba Formation sandstones have shallow, leached, acidic sandy soils (Mucina et al., Reference Mucina, Scott-Shaw, Rutherford, Camp, Matthews, Powrie, Hoare, Mucina and Rutherford2006) that are dominated by sourveld grasslands (Pondoland-Ugu Sandstone Coastal Sourveld). Thicket vegetation (i.e., Eastern Valley Bushveld) is common in fire-resistant areas like riverine floodplains. Forests are found throughout the area and include tall, species-rich, and structurally diverse subtropical Scarp Forests at the coast, along cliffs, and in gorges; Southern Coastal Forests along the coastline that are dominated by Celtis africana and Sideroxylon inerme (milkwood tree); and Southern Mistbelt Forests further inland at altitudes between 850 and 1100 m above sea level (Mucina and Geldenhuys, Reference Mucina, Geldenhuys, Mucina and Rutherford2006) (see Fig. 2). The resilience of Scarp Forests and Afrotemperate Forests in the Eastern Cape Province and KwaZulu-Natal has made these habitats important ecosystems during glacial phases and, particularly, during the LGM (Lawes et al., Reference Lawes, Eeley, Findlay and Forbes2007; Esteban et al., Reference Esteban, Bamford, Miller, Neumann, Schefuß, Zabel, Pargeter, Cawthra and Fisher2019).

Prior archaeological research in eastern Pondoland

The ~100 km stretch of Eastern Cape coast between Port St. Johns in the South and the Mtamvuna River in the north preserves a rich archaeological record that dates back ca. 300 ka (Fisher et al., Reference Fisher, Albert, Botha, Cawthra, Esteban, Harris and Jacobs2013; Fisher, Reference Fisher2016). Yet almost nothing was known about Pondoland's pre–Iron Age archaeological record before the inception of our research project. Chubb and colleagues (Reference Chubb, King and Mogg1934) documented excavations at a local rock shelter in 1932, and Davies reported on undated surface lithic scatters from the broader Natal region, which he attributed to the Earlier Stone Age's “Sangoan” Industry (Davies, Reference Davies1976, Reference Davies1982). Kuman and colleagues briefly report on extensive surface exposures of handaxes and other large bifacial tools along northern Pondoland's Red Sands area at Xolobeni (Kuman and Clarke, Reference Kuman and Clarke2005). But apart from these preliminary reports, and the research by Feely and colleagues (Feely, Reference Feely1980, Reference Feely1985, Reference Feely1986; Granger et al., Reference Granger, Hall, McKenzie and Feely1985; Feely and Bell-Cross, Reference Feely and Bell-Cross2011) that focused on Iron Age occupations, there are no other published accounts about the pre–Iron Age prehistory from the area. What little is known comes from sites in the surrounding region—Sibudu Cave, Border Cave, Sehonghong, Strathlahan, and Umhlatuzana—yet each of these sites is at least 350 km away from eastern Pondoland and none are coastal (see Fig. 1).

Importantly, Pondoland and the broader Eastern Cape provides observations of some of southern Africa's only historically documented mountain and coastal hunter-gatherer groups. Historical reference to hunter-gatherer groups in the Eastern Cape mention ephemeral open-air hunting camps (Vinnicombe, Reference Vinnicombe1976:30). Historical records also suggest that some recent hunter-gatherer groups partook in short-term seasonal logistical forays from the nearby Maloti-Drakensberg Mountains into lowland areas of KwaZulu-Natal and the coastal lowlands of the Eastern Cape to hunt, acquire raw materials not available in the highlands, and engage in trade and stock theft from neighboring farmers (Wright, Reference Wright1971). For example, the widely used plant poisons derived from Acokanthera oblongifolia and Acokanthera oppositifolia, which were common to highland hunter-gatherer groups, are species not known to occur in the Lesotho highlands but are prevalent in the vegetation of the Eastern Cape and KwaZulu-Natal Provinces (Cable, Reference Cable1984). Acokanthera poisons are also made more toxic during the drier winter seasons, suggesting possible seasonal mobility between interior highland and coastal lowland regions (Schapera, Reference Schapera1923).

In Africa, ethnographic examples of logistical mobility—the movement of resources to consumers, sensu Binford (Reference Binford1980)—are rare, and Marean (Reference Marean2016) has argued that the African archaeological record is likely to comprise a “modern human hunter–gatherer adaptive system typified by high residential mobility.” However, logistical mobility may have been more widespread than can be inferred from existing ethnographic evidence, which has been derived mainly from resource-poor environments like the Kalahari that are poorly suited to the use of logistical mobility strategies. In the cross-section between the Lesotho escarpment and the Eastern Cape coastal forelands, the available ethnographic records of recent hunter-gatherers hints at complex seasonal mobility patterns, which may themselves reflect a pattern that existed earlier in the Holocene and even the late Pleistocene. The presence of a single marine shell bead (Nassarius kraussianus) derived from early LGM (23,800–24,400 cal yr BP) layer BAS at Sehonghong rock shelter in Lesotho, for example, as well as numerous marine shell beads from Indian Ocean contexts in Holocene deposits certainly support this inference (Mitchell, Reference Mitchell1996). These patterns are associated with seasonal food scarcities and the cyclical availability of resources. Few of Africa's other ethnographic archives describe hunter-gatherers living in such conditions. A more detailed examination of Pondoland's archaeological traces provides insight on a matter currently under-described in the ethnographic record.

The P5 Project

The Pondoland Paleoclimate, Paleoenvironment, Paleoecology, and Paleoanthropology Project (P5) is an interdisciplinary and international collaboration of researchers focused on understanding human adaptations to coastal environments. Early work in Pondoland by P5 focused on identifying and describing coastal archaeological records in rock shelters and open-air contexts. In 2011, P5 surveyed the Mkambati Game Reserve and the Msikaba and Lambasi areas of eastern Pondoland, documenting 20 rock shelters and 5 open-air sites, most having abundant marine mollusk shells, stone tools, and terrestrial faunal remains. These surveys provided the first modern archaeological inventory of coastal occupations in Pondoland, spanning the Early Stone Age to historic European shipwrecks. Full details of the survey and sites are provided in Fisher and colleagues (Reference Fisher, Albert, Botha, Cawthra, Esteban, Harris and Jacobs2013).

In 2015, P5 conducted test excavations at four previously documented open-air and rock-shelter sites within the study area (Fisher, Reference Fisher2016). One of these sites is a large, dry coastal rock shelter (~80 m × 20 m) located ~25 meters above mean sea level (amsl). Originally designated using its grid coordinates as Site A2SE-1 (i.e., the first site recorded within the SE sub-quadrant of survey grid square A2), the site is now generally referred to as “Waterfall Bluff,” and that name is used here (see Fisher et al., Reference Fisher, Albert, Botha, Cawthra, Esteban, Harris and Jacobs2013 for details about the survey grid). The test excavations at Waterfall Bluff revealed well-stratified archaeological deposits with abundant stone artifacts, shell, well-preserved bone, and macrobotanical remains. In 2016, P5 expanded the excavation at Waterfall Bluff to 4 m × 1 m (Fig. 3). These excavations revealed a well-preserved stratigraphic sequence that exhibits a north-south gradient of stratigraphic preservation from the dripline (poorer preservation) to more protected spaces of the shelter (better preservation) (Fig. 4). Continued excavations at Waterfall Bluff since 2018 have exposed a larger sample of lower deposits at the site confirming better preservation of stratigraphy further inside the rock shelter.

Figure 3. (color online) Planimetric map of Waterfall Bluff. This image shows a map of the site and excavation areas as well as key geomorphic features like the dripline and overhead cliff face. An inset showing the main excavation area is also provided, as is a cross section of the cliff face and rock shelter at A–A’. UTM 35S northing and easting coordinates in meters are provided along the north (vertical) and east (horizontal) axes.

Figure 4. The stratigraphic sequence from the main excavation area at Waterfall Bluff. This image shows the E321.5 profile (left), the N850.5 profile (center) and the E321 profile (right), which are adjacent and wrap around three-quarters of the main excavation area at Waterfall Bluff. At top is the color-corrected, high-resolution, georectified photomosaic of the sections that was created from >300 individual photographs. SubAggregate boundaries have been superimposed on this image, and the lower image shows the SubAggregates more clearly. A color key reference of the SubAggregates is provided below. The vertical axis represents orthometric height in meters, and the horizontal axis represents truncated UTM35S northing and easting coordinates. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Excavation methodology

The P5 excavation methodology is adapted from protocols developed by the South African Coast Paleoclimate, Paleoenvironment, Paleoecology, and Paleoanthropology (SACP4) project for 3D mapping and record keeping (Bernatchez and Marean, Reference Bernatchez and Marean2011; Oestmo and Marean, Reference Oestmo, Marean and Smith2014). In preparation for excavations, we established geodetic control at the site using a Topcon Hyper XT RTK Global Navigation Satellite System (GNSS) system to survey permanent control points in front of the rock shelter within the Universal Transverse Mercator (UTM) coordinate system (Zone 35S, World Geodetic System [WGS] 1984). GNSS static station measurements were recorded in 1-second intervals over 3 hours to account for ~100 km baseline lengths to the nearest TrigNet stations at Kokstad and Mthatha. Static station vectors were calculated to each TrigNet station using Carlson Survey GNSS 2016 software, and a least-squares adjustment was used to correct the rover data. Fixed L1 solutions with 100% ambiguity fixing was achieved for each point. The maximum 3D position of the mean dilution of precision (PDOP) was 2.2 with a mean PDOP of 1.8. Ellipsoidal WGS84 elevations were then converted to orthometric heights using SAGEOID10 (Chandler and Merry, Reference Chandler and Merry2010). We used these control points and a Topcon ES-105 reflectorless total station to traverse into the rock shelter and survey additional control points on the shelter walls. The control points inside the rock shelter now enable our total stations to be set up anywhere within the rock shelter and surrounding landscape and be tied into the global UTM system for precise and accurate digital measurements.

Using the UTM system, we have gridded Waterfall Bluff into 1-m squares where each square is named using the last three digits of the UTM Easting and Northing coordinate of the southwest (SW) corner of the square. Furthermore, excavations are conducted within 50 cm × 50 cm quadrants within each square, designated as SW, northwest (NW), northeast (NE), or SE. Therefore, N854E321SE refers to the SE quadrant within grid square N854E321, itself having SW corner coordinates Northing 6518854.0 and Easting 768321.0.

All excavations follow natural stratigraphy, and our tiered classification of the stratigraphy follows the methodology of Karkanas and colleagues (Reference Karkanas, Brown, Fisher, Jacobs and Marean2015). Accordingly, stratigraphic layers (i.e., “StratUnits”) represent anthropogenic and geogenic changes in sediments that are visible during excavations. StratUnits are grouped into larger “SubAggregates” (i.e., “SubAggs”), which represent identifiable and discrete anthropogenic, biogenic, or geologic events. A cluster of shell-rich deposits with interdigitated fine sterile layers, for example, may be classified as a SubAgg if it can be shown that these layers were deposited contemporaneously and under similar processes. SubAggs are then consolidated into “Stratigraphic Aggregates” (i.e., “StratAggs”), that are laterally continuous sediments across large areas. These sediment groups represent periods of specific formation processes throughout a site's life history. StratUnits are identified using an alphanumeric system. StratUnit WB16AC, for example, refers to the 29^th StratUnit assigned at Waterfall Bluff during the 2016 field season (i.e., 26 StratUnits from A to Z, then cycling to AA, AB, AC, etc. for StratUnits 27 and so on). SubAggs are assigned proper names associated with the project. StratAggs are identified using an acronym indicating the broad characteristic of the sedimentological group, like SRCS, which means “Shell-Rich Clayey Sands.” Throughout this paper, we refer to all deposits and archaeological materials with reference to StratAggs and SubAggs, descriptions of which are provided in Table 1 and Figure 4.

Table 1: Table summarizing the SubAggregate stratigraphy at Waterfall Bluff. This table provides details about the stratigraphy at Waterfall Bluff in chrono-stratigraphic order whenever possible. Black bars are used to represent non-contiguous contacts for spatially isolated layers within the overall chrono-stratigraphic sequence. The 68.2% modeled radiocarbon boundary ages (see Table 2) and 1-sigma optically stimulated luminescence (OSL) results (Table 3) for SubAggs are provided as a reference when available.

Additionally, each StratUnit is assigned a unique “Lot Number,” which is a bulk tracking number. It is assigned to all artifacts, fauna, or features that are mapped and/or collected within a single quadrant of a single StratUnit. All 3D-plotted archaeological materials and fauna are further assigned unique Catalog Numbers (CN). References to specific artifacts or samples always include the Lot and CNs (e.g., CN123456, Lot 123), which identify the unique ID of every object but also constitute a direct link to full contextual information about the object's provenience in our Microsoft Access tabular records and in our 2D and 3D Geographic Information System databases.

All sediments removed during excavations are wet-sieved through nested 5-mm, 3-mm, and 1.5-mm screens before the remainder is dried and bagged individually using the Lot Number, StratUnit, and mesh size to identify the bag's provenience and contents. In the P5 laboratory at the East London Museum, these screened materials are sorted into different categories (e.g., shell, stone, fauna, etc.), and all identified screened materials are assigned their own CNs and placed into individual plastic bags.

Except for during our earliest test excavations, we plot all archaeological remains (including, but not limited to artifacts, charcoal, ochre, macro-plants, terrestrial and marine fauna, and shellfish) and samples for specialized analyses (e.g., geochronology, geochemistry, palynology, phytoliths, plant wax isotopes, and magnetic susceptibility) in 3D with a total station. We also collect sediment samples from each StratUnit—known as a “Bulk Sample”—for any additional analyses that might be found useful in the future. Every rock >5 cm found during excavations is also mapped, as is every StratUnit surface. At the end of excavations, we use a Nikon D300s and 35 mm f1.8 lens to create high resolution (~1 mm pixel resolution), color-corrected photomosaics of all stratigraphic profiles that are georectified in 3D. These methods are described in Fisher and colleagues (Reference Fisher, Akkaynak, Harris, Herries, Jacobs, Karkanas, Marean and McGrath2015), except that Imatest 5.1 is now used to color correct the imagery.

The site and surrounding cliffs have been mapped using a combination of total stations (Topcon ES-Series and Northwest NTS02S), RTK GNSS (Topcon Hyper XT and Carlson BrX5), and forward-facing georectified parallactic photography via a DJI Mavic 2 Pro and a Phantom 2 mounted with a GoPro Hero 3 + . These data, and the excavated point-plot data, have all been integrated into a high-resolution and photorealistic 3D Geographic Information Systems model using ArcGIS Pro 2.1.3, and to a lesser extent ESRI ArcGIS 10.6. The models allow our team to make detailed observations about the rock shelter, excavated archaeological deposits, cliffs, and surrounding coastal shelf to understand site formation processes and, ultimately, the context of the archaeological deposits.

Excavation history

Excavations began at Waterfall Bluff in 2015 with the excavation of two 50-×-50-cm test pits. One test pit (N844E335SW) was located in the southeastern part of the rock shelter whereas the other (N854E321NW) was located near the central/NW part of the rock shelter. Both excavations were located within the dripline and on the gently sloping surface that backs up against the rear of the site.

Quad N844E335SW reached a maximum depth of ~60 cm. Excavations here ended prematurely on account of time, but they revealed a well-stratified archaeological sequence. Observations of the east- and west-facing profiles within quad N844E335SW suggest that the contemporary 10° slope of the rock shelter surface may be due to erosion of deposits underlying modern overburden sediments (Fig. 5).

Figure 5. (color online) Stratigraphic profiles from quad N844E335SW. Quad N844E335SW is located in the southeast section of the rock shelter at Waterfall Bluff. The sequence shows a coarse sandy layer (Light Brown Coarse Sands [LBCS] Siyambonga) that is similar to other LBCS sediments elsewhere on site. Above this, the sedimentology is more clayey and reminiscent of the Shell-Rich Clayey Sands (SRCS). Two SRCS SubAggregates have been identified, charcoal-rich SRCS Josh and shell-rich SRCS Nate.

Quad N854E321NW, in the center of the shelter, reached a maximum depth of ~45 cm before reaching a large, flat stone surface that we now know to be a fallen block from the roof. In 2016, we expanded the excavations around quad N854E321NW and were able to excavate around the roof block and remove it, revealing more archaeological deposits underneath. The excavation area was also expanded laterally to 4.5 meters long × 0.5 meters wide on a north–south axis (see Fig. 4). The location of the 4.5-m-long trench was designed to intersect the dripline at the south end of the excavations (i.e., along the N850.5 line) and provide information on sedimentological and archaeological preservation in the shelter.

The deepest current excavation areas at Waterfall Bluff are located at the dripline in quad N850E321NW (see Fig. 4). Here, we excavated to ~80 cm below the modern shelter surface. These excavations revealed stratified archaeological layers and a major change in sediments representing two distinct StratAggs, which are described below. Bedrock has not yet been reached.

During the course of excavations, we uncovered a recent trash pit, ~80 cm wide × ~50 cm deep, that had been dug into archaeological deposits between N853 and N854 and that was partially burned in situ (see Fig. 3). Using a Cadbury Smash instant mashed potato wrapper found within the deposit, we were able to relatively date the trash pit to the mid–late 1970s. We opted to excavate the trash pit en masse, which exposed an ~1-m-wide pit that we continued to excavate to expose underlying in situ archaeological deposits.

In 2019, significant efforts were made to deepen the main excavation trench to collect a larger sample of archaeological materials from SRCS and Light Brown Coarse Sands (LBCS) contexts and to provide more information about the stratigraphy. These excavations have revealed a complex stratigraphic sequence, which shows evidence for not only vertical accretion of deposits, but also lateral accretion from the rear of the rock shelter toward the mouth. Stratigraphic preservation does, therefore, appear to increase within the confines of the shelter, and many SubAggs can now be traced across most of the excavated trench (see Fig. 4).

Stratigraphic sequence

We have subdivided the current excavated deposits into two StratAggs, named the LBCS and the SRCS. Currently, the deepest StratAgg is the LBCS. Figure 4 shows the primary stratigraphic profiles at the site as well as the locations of each SubAgg documented to date. Additionally, Table 1 summarizes the sedimentology and archaeology of each SubAgg and provides comments regarding how these layers are currently interpreted. A detailed micromorphological investigation is under way.

In general, the LBCS is characterized by loosely compacted moist coarse sands. Interspersed within these sands are darker deposits having relatively greater archaeological remains, ash, and charcoal. None of the deposits completely lack archaeological remains. This suggests that the site was never abandoned for prolonged periods of time during the LGM like sites in the surrounding highland and inland areas (Stewart and Mitchell, Reference Stewart and Mitchell2018). Instead, the changes in archaeological content across the LBCS may signal fluctuations in occupation intensity or differential use of the site over time, though aspects of duration or periodicity are currently unknown.

Multi-proxy evidence of marine shellfish and marine fish remains have now been recovered from multiple LBCS layers attesting to the proximity of the coastline during this time. At the dripline, this evidence ranges from microscopic to macroscopic. For example, scanning electron microscopy (SEM) and X-ray diffraction (XRD) have been used to identify microscopic traces of dissolved shellfish in LBCS Kuka representing mytilid mussel (probably the locally dominant Perna. perna) based on the microstructure of the shell (Fig. 6). Subsequent investigations have also revealed marine shell fragments and marine fish bone, which are relatively better preserved from LBCS layers found inside the rock shelter, though most of these materials remain under investigation and are not reported here.

Figure 6. Scanning electron microscopy (SEM) of shellfish. SEM nacre tablets present in sediment sample 315 Kuka (a–d). Images (c) and (d) show the nacre tablets formed as sheets, which match with the modern mussel (P. perna) samples (e, f) that highlight the interior aragonitic sheet nacreous microstructural layer. SEM images were taken with various settings due to a high occurrence of beam charging—seen in images (c) and (d)—despite sputter coatings of carbon and gold. (a) and (b) 10Kv backscatter electron detector. (c) and (d) 10kv secondary electron detector (SED). (e) 15Kv SED. (f) 20Kv SED.

Stratified above the LBCS, the SRCS is sedimentologically distinct, having a dark, clay-rich composition nearly lacking in roof spall. The SRCS stratigraphy is complex, and there are several instances where SRCS deposits have been truncated and modified by prehistoric site inhabitants contemporaneously with their formation. One of the clearest examples of this kind of site modification is found in SRCS Bonile, which was truncated between N853.5 and N854 soon after deposition to make room for the deposition of the SRCS Lwando midden (see Fig. 4). SRCS Lwando itself is composed of predominantly burnt shells, yet there is little evidence of in situ combustion features, which suggests the Lwando shells are in a secondary context and were dumped in the hole created from the removal of SRCS Bonile after they had been burned.

The same sequence of StratAggs is also seen in N844E335SW (see Fig. 5). The test excavation is ~15 meters SW of the main trench. The current lowest deposit is an undated roofspall-rich coarse-grain sand called LBCS Siyambonga. This deposit is overlain by loosely consolidated dark and organic-rich sands, which have been subdivided into two SubAggs, SRCS Josh (lower) and SRCS Nate (upper).

Chronology Methods

AMS ¹⁴C Dating Methods

Fifty-one AMS ¹⁴C samples have been dated from our catalog of piece-plotted wood charcoal. The proveniences of all samples are listed in Supplementary File 2. Charcoal wood identification of the samples is currently under way. Most of the charcoal was piece-plotted in situ and put into individual sterile plastic containers immediately upon discovery to limit contamination.

The initial radiocarbon analysis was conducted by Beta Analytic, Inc, and these samples are reported in Supplementary File 3. Using an acid-alkali-acid pretreatment procedure, these samples were first floated and agitated in deionized H₂O to disperse the sediment. They were then progressively sieved through 250- and 180-μm-diameter sieves to remove any plant macrofossils or rootlet material prior to pretreatment. The <180 μm fraction was then subjected to a series of hot (near boiling) 0.5N HCl leaches to remove any carbonates before being immersed in an alkali wash (NaOH), which removed secondary organic acids. The samples were then rinsed with deionized H₂O until neutral and dried in a 70°C oven overnight. The pretreated sample material was then homogenized and examined under a 45× microscope to inspect for any rootlet hairs or fragments. A portion of each sample was tested with 0.5N HCl again to ensure that all carbonate had been removed. The sample was then combusted to CO₂ and graphitized for AMS counting.

P5 has subsequently partnered with iThemba Laboratory for Accelerator Based Sciences (iThemba LABS) to promote their AMS dating facility, which is currently the only one operating in Africa. Charcoal sub-samples measured at iThemba LABS were pre-treated with a standard acid-base-acid treatment: 1% HCl, 1% NaOH, and 1%HCl, respectively. Each treatment step was carried out for 45 minutes at 70°C. Sub-aliquots of the pre-treated sample were analyzed for δ¹³C at the stable isotope facility of the University of Pretoria.

For ¹⁴C analysis, the pre-treated charcoal samples were graphitized at the iThemba LABS graphitization line and then pressed as sputter target for AMS analysis. The AMS system at iThemba LABS is based on a 6MV EN tandem multi-sample Cs-sputter ion source (similar to the models used by Lawrence Livermore National Laboratory and PrimeLAB), a multi-beam switching system capable of fast cycling between isotopes (run at 5 cycles per second), and a modern AMS analysis beam-line supplied by NEC (Mbele et al., Reference Mbele, Mullins, Winkler and Woodborne2017). At the accelerator we use ¹³C for normalization, and δ¹³C correction is based on the conventional mass spectrometry measurements. Samples were normalized and blank corrected against graphitized samples of the OXII standard and our fossil coal blank material (the latter pre-treated with the same protocol as the charcoal), respectively. In a measurement run, up to 43 unknown samples are analyzed together with a set of at least 5 OXII standard samples for normalization. In addition, 6 blanks are routinely measured in every such run; however, a long-term averaged blank value and the corresponding long-term standard deviation is used for the correction. The samples in the ion source cycled through several times to check against possible long-term drifts. Together with fast-cycling and the use of “bad-cycle” rejection, we can thus eliminate both short-term and long-term conditions not conducive to high reproducibility. AMS data analysis follows the concepts put down by Puchegger and colleagues (Reference Puchegger, Rom and Steier2000).

Additionally, a previous set of charcoal samples that were measured by Beta Analytic, Inc. and that still had material available was subsequently measured at iThemba LABS. The results between the laboratories for these samples agree within error. A set of samples have also been run in duplicate at iThemba LABS to affirm overall sample pre-treatment and AMS inter-run reproducibility. The individual AMS radiocarbon results for iThemba LABS measurements and combined values for the charcoal samples are available in Supplementary File 3.

The Bayesian radiocarbon age model presented herein is based on 51 radiocarbon dates of charcoal samples for 13 SubAggs, using OxCal v4.3 (Ramsey, Reference Ramsey2009a). SHCal13 (Hogg et al., Reference Hogg, Hua, Blackwell, Niu, Buck, Guilderson and Heaton2016) was used for calibration. The model parameters are provided in Supplementary File 4. The raw data from the model are provided in Table 2. The ages that are used here to describe each SubAgg, including in Table 1, are based on data at the 68.2% confidence interval rounded to nearest decade between 10,000 and 25,000 BP; ages greater than 25,000 BP are rounded to the nearest half-century. This rounding prevents over-reporting of significant digits while ensuring that any shift thereby introduced stays below a sixth of the 68.2% range so as to be irrelevant. Lastly, Figure 7 provides a graphic illustration of the posterior density function and range and the unmodeled probability density function and range for each sample and SubAgg boundary at the 68.2% range.

Figure 7. Bayesian model results. This figure shows the results of a Markov Chain Monte Carlo analysis of the radiocarbon data from Waterfall Bluff at both 68.2% and 95.4% ranges. The samples are ordered in chrono-stratigraphic order. For all samples, the posterior density function and range are represented in red, and the unmodeled probability density function and range are represented in gray. The ranges of Marine Isotope Stage (MIS) 3, MIS 2, and the Early Holocene are demarcated with vertical gray bars, as are the last glacial maximum (LGM) and last glacial/interglacial transition (LGIT), showing which samples date to these time periods. Overall, the chronological sequence shows persistent occupation of the rock shelter from Late MIS 3 to the Early Holocene, including during the LGM and LGIT. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 2. AMS 14C Bayesian Analysis Results. This table provides the unmodeled and modeled results of a Markov Chain Monte Carlo analysis of the radiocarbon data from Waterfall Bluff at both 68.2% and 95.4% ranges. The SHCAL13 calibration curve (Hogg et al., Reference Hogg, Hua, Blackwell, Niu, Buck, Guilderson and Heaton2016) was used. Results >25,000 BP were rounded to the nearest 50 years, and results between 10,000 and 25,000 years were rounded to the nearest 10 years to prevent over-reporting of significant digits.

Based on the species analysis for the charcoal (ongoing and not presented here) we anticipate no significant old-wood effect. The SubAggs LBCS Kate and LBCS Hayley are modeled as potentially overlapping, and currently we do not have a radiocarbon date for LBCS Colton. The current excavated limits of LBCS Hayley do not overlap with LBCS Otis; therefore, it cannot provide a constraint on LBCS Otis or vice versa (see Fig. 4).

We have a sequence of deposition from LBCS Otis to SRCS Justin. The choice of contiguous boundary conditions from SRCS Courtney through to SRCS Justin is partly informed by the fact that any attempt to make them sequential (potentially allowing for a gap of deposition) does not yield any clearly evident gaps in deposition—all of them are consistent with 0 on both the 95.4% and 68.2% confidence intervals and are certainly below the resolution of radiocarbon measurements themselves. The sequential or contiguous boundary conditions elsewhere are informed by the evidence from the interfaces between the SubAggs. Specifically, we assume sequential boundary conditions between SRCS Courtney–LBCS Kuka, LBCS Kuka–LBCS Chopi, and LBCS Chopi–LBCS Lily, but contiguous boundary conditions between LBCS Lily and LBCS Otis. Of all the potential gaps in this specific sequence, only the one between LBCS Chopi and LBCS Lily could be considered substantially evident, but there is evidence to suggest this may be due to sampling strategy.

Modelling of the SubAggs within SRCS also proved problematic as the calibration curve intersects 3 times over a 500-year period for all SRCS results. Potentially small deviations, therefore, could have significant consequences around that time period and tip the whole set of SubAggs to the older or younger side of the wiggle. Four samples from this period show low agreement for the model; however, removing them did not have a significant impact on the modelling result, and they are therefore included. Furthermore, OxCal runs using a simple outlier model (Ramsey, Reference Ramsey2009b) also did not conclusively suggest elimination. Thus, there is good indication that SRCS Courtney is on the older side while SRCS Lwando, Jess, and Justin are highly likely to be on the younger side, but the exact transition from SRCS Lwando to SRCS Courtney is highly uncertain and made difficult by SRCS Bonile having only one radiocarbon result.

OSL Dating Methods

OSL dating provides an estimate of the amount of time elapsed since mineral grains were last exposed to sufficient sunlight (Huntley et al., Reference Huntley, Godfrey-Smith and Thewalt1985; Aitken, Reference Aitken1998; Duller, Reference Duller2004; Wintle, Reference Wintle, Holland and Turekian2014; Roberts et al., Reference Roberts, Jacobs, Li, Jankowski, Cunningham and Rosenfeld2015). This method can be used to determine the depositional age of sediments and, by association, the age of material remains at archaeological sites. Four sediment samples were collected from the LBCS deposits, and one was collected from the transition between LBCS and SRCS. Sand-sized quartz grains were extracted using standard sample preparation procedures (Wintle, Reference Wintle1997; Aitken, Reference Aitken1998). Equivalent dose (D_e) values were estimated for single grains using the single aliquot regenerative-dose procedure (Galbraith et al., Reference Galbraith, Roberts, Laslett, Yoshida and Olley1999; Murray and Wintle, Reference Murray and Wintle2000). Details about sample preparation, equipment, measurement, and data analysis procedures are provided in Supplementary File 5. Grains with aberrant OSL properties were rejected from final D_e determination (Jacobs et al., Reference Jacobs, Wintle, Roberts and Duller2008), following the rejection criteria outlined in Supplementary Table S5-2. Environmental dose rates were calculated as the sum of the external beta, gamma, and cosmic-ray dose rates and a small alpha dose-rate contribution from inside the mineral grains.

The accepted D_e values for individual grains from each sample are displayed on radial plots in Supplemental Figure S5-3. If the individual values are consistent with statistical expectations, then 95% of the points should scatter within a band ± 2 units wide, projecting from the “standardized estimate” axis on the left to any chosen D_e value on the radial axis to the right (Galbraith et al., Reference Galbraith, Roberts, Laslett, Yoshida and Olley1999). It is apparent from the plots for each sample that the D_e estimates are spread widely and do not fall within a single band of ± 2 units. This spread is also indicated in the D_e overdispersion (OD) values (Table 3), which range from 37 ± 3% (CN311) to 50 ± 6% (CN312), and are much greater than the 14 ± 4% OD value obtained for sample CN308 under controlled laboratory conditions in a dose recovery test (Supplemental Fig. S5-2). Two types of D_e distributions were observed for the Waterfall Bluff samples. The first type (sample CN311) shows a broad distribution that is scattered around a central value (Supplemental Fig. S5-3a). For this sample, we used the central age model (CAM) (Galbraith et al., Reference Galbraith, Roberts, Laslett, Yoshida and Olley1999) to obtain a best-estimate final D_e value for age determination. The CAM takes into account the OD; the greater the OD, the larger the associated uncertainty on D_e. The second type (samples CN308, CN309, and CN312) also shows a broad distribution, but with a number of additional characteristics.

Table 3. Final environmental dose rate results, D_e values, and optically stimulated luminescence (OSL) ages for Waterfall Bluff samples. D_e and OSL age results using the 3-parameter minimum age model (MAM) (Galbraith et al., Reference Galbraith, Roberts, Laslett, Yoshida and Olley1999) are given in addition to the central age model (CAM) (Galbraith et al., Reference Galbraith, Roberts, Laslett, Yoshida and Olley1999) for samples CN308, CN309, CN310, and CN312. MAM OSL age determinations for these samples are currently assumed to be most representative of sample burial age but require further investigation.

a Estimated water content over period of burial of sample. Current measured field water content value given in brackets.

b Overdispersion (OD) values calculated using the CAM.

First, some grains have much higher than expected D_e values (up to ~225 Gy). These high D_e values form part of a distribution of D_e values that is truncated at the upper end of the distribution (Supplemental Fig. S5-3c–e), excluding grains saturated with respect to dose (criteria 7 and 8 in Supplementary Table S5-2) from which no D_e value could be calculated. This means that the calculated OD values are minimum values only, and that the true spread in D_e values is greater than that observed in the radial plots.

Second, these samples also show no evidence for the presence of discrete D_e components. Instead, these samples have continuous D_e distributions that can be interpreted as evidence for either substantial bioturbation of sediments or partial bleaching of sediment, including the liberation of grains from roof spall. Sample CN310 has “transitional” D_e characteristics between these two types of distributions. For samples CN313, CN310, CN309, and CN308, further work on understanding site formation and sediment context at the microscopic level is required to make an informative interpretation of the D_e distributions and to help choose an appropriate age model for final D_e determination. In the absence of this information, we calculated both the CAM D_e value and a minimum D_e value, using the 3-parameter minimum age model (MAM) of Galbraith and colleagues (Reference Galbraith, Roberts, Laslett, Yoshida and Olley1999) for each of these samples (Table 3).

We calculated the beta and gamma dose rates for each of the samples using two different approaches. In one approach, we converted concentrations of U, Th, and K obtained from inductively coupled plasma mass spectrometry (ICP-MS) or optical emission spectroscopy (ICP-OES) to beta and gamma dose rates. In the other approach, we directly measured the beta dose rates using GM-25-5 beta counting (GMBC; Bøtter-Jensen and Mejdahl, Reference Bøtter-Jensen and Mejdahl1988) and calculated the gamma dose rate from U and Th obtained through thick source alpha counting (TSAC) and K from a combination of GMBC and TSAC. Both approaches gave statistically consistent beta and gamma dose rates (average beta dose rate ratio = 0.97 ± 0.05 and average gamma dose rate ratio = 1.02 ± 0.04; Supplementary Tables S5-5, S5-6). For final age determination, we used the dose rate values derived from ICP-MS/ICP-OES (Table 3). There is a significant difference between the dose rates obtained for the samples from the shell- and clay-rich StratAgg SRCS (SubAgg Courtney) and the underlying LBCS sediments.

The final D_e values, environmental dose rates, and OSL ages for all five samples are listed in Table 3. To obtain an age for each sample, the model-derived D_e value was divided by the total environmental dose rate. For samples CN308, CN309, CN310, and CN312, the CAM D_e values result in significant overestimation of age when compared to AMS ¹⁴C ages. For these samples, the MAM D_e values instead gave ages that are in better agreement with those obtained by AMS ¹⁴C dating, suggesting that the sediment may have been partially bleached prior to deposition and burial. We caution that we need further information about the sedimentary context and site formation to be confident in the OSL ages for these sediments. Uncertainties on reported ages are given at the 1-sigma level.

Chronology results

OSL dating (Table 3) and Bayesian ¹⁴C (see Table 2) modeled results are summarized below. The minimum age of sediment deposition for the currently deepest excavated deposits at Waterfall Bluff (LBCS Colton) have been OSL dated to 37.6 ± 4.2 ka (CN312). Bayesian ¹⁴C model results for the start of overlying LBCS Kate range from 32,000 and 31,050 cal yr BP and end between 31,000 and 30,100 cal yr BP. These boundary age estimates are consistent with the minimum age of sediment deposition for LBCS Kate, OSL dated to 27.8 ± 2.1 ka (CN308). Currently, no samples have been dated from LBCS Ollie, which directly overlays LBCS Kate. However, the next most recent deposit chronologically is LBCS Hayley, which has a lower Bayesian ¹⁴C model boundary result from 28,150 to 27,150 cal yr BP and an upper boundary from 24,050 to 23,160 cal yr BP.

Stratigraphically above LBCS Ollie is LBCS Otis. Bayesian ¹⁴C model results of the start of LBCS Otis range from 23,080 to 22,610 cal yr BP and end between 22,560 and 21,910 cal yr BP. These results are within the error range of the minimum age of sediment deposition for LBCS Otis, OSL dated to 22.1 ± 1.3 ka (CN309).

Bayesian ¹⁴C model results for LBCS Lily, which overlays LBCS Otis, range from 22,560 to 21,910 cal yr BP and end between 20,190 and 19,330 cal yr BP. LBCS Chopi deposits are not laterally extensive but do overlay LBCS Lily. Bayesian ¹⁴C model results for LBCS Chopi range from 18,080 to 17,250 cal BP and end between 15,050 and 14,440 cal yr BP.

LBCS Kuka shares a lower boundary with LBCS Lily, LBCS Chopi, and LBCS Hayley deposits, underscoring the complex and spatially discrete distribution of the stratigraphy across the site. LBCS Kuka is also the uppermost SubAgg of the LBCS, and its upper boundary marks the transition to SRCS where it shares a boundary primarily with LBCS Courtney, but also SRCS Irene. Bayesian ¹⁴C model results for LBCS Kuka interpret the lower boundary to date from 14,070 to 13,570 cal yr BP and the upper boundary from 13,520 to 12,830 cal yr BP. This is consistent with the minimum age of sediment deposition given by CN310, OSL dated to 12.5 ± 1.2 ka.

SubAgg Courtney is the stratigraphically deepest layer of the SRCS. Bayesian ¹⁴C model results date the lower SRCS Courtney boundary from 11,160 to 11,050 cal yr BP and the upper boundary between 11,060 and 10,990 cal yr BP. The OSL age of sample CN311 is slightly older than these results at 13.3 ± 0.7 ka, a minor discrepancy that can be attributed to an increased potential for underestimation in beta dose rate calculations (and thus OSL age overestimation), caused by the shell-rich deposit (Cunningham, Reference Cunningham2016).

Bayesian ¹⁴C model results for overlying SRCS layers have been calculated (in chronostratigraphic order) for SRCS Irene, Irit, Bonile, Lwando, Jess, and Justin. From SRCS Courtney to SRCS Justin, the Bayesian ¹⁴C model results calculate that these deposits formed in less than 450 years.

A later SRCS sequence dating to the Middle Holocene has been discovered in N844E335SW. AMS ¹⁴C dates from SRCS Josh and SRCS Nate have been dated to 8,395–8,203 cal yr BP and 8,288–8,025 cal yr BP, respectively. Since the relationship between these deposits and those in the main excavation area remain unknown, they were not used in the Bayesian model.

Lithics

In total, 403 lithics have been analyzed from the LBCS and SRCS to date. These data draw primarily on plotted cores and flakes, omitting plotted flakes <5 mm in size and all sieved materials. The results provided here, therefore, should be considered preliminary given this limited sample and the absence of fine-mesh-sieved materials in the current analysis. We used a standardized attribute-based framework focused on variables to clarify strategies in raw material selection, reduction strategies, and techniques. The variables also help clarify two important aspects of lithic technology—the organization of technology and the size and shape of flakes. The LBCS sample comes predominantly from SubAggs Kuka, Lily, and Otis, while the SRCS sample draws largely from Bonile, Courtney, Jess, and Rosa. Table 4 summarizes the lithic data by class and raw material within both the LBCS and SRCS. Table 5 summarizes the proportions of cores versus flakes within the analyzed SubAggs. We have yet to analyze lithics from the deepest excavated deposits in LBCS Colton.

Table 4. Summary of Shell-Rich Clayey Sands (SRCS) and Light Brown Coarse Sands (LBCS) lithics. Counts and percentages of different lithic types and their raw materials within the two stratigraphic aggregates at Waterfall Bluff.

Table 5. Cores and Flakes in SRCS and LBCS. Distribution of cores and flakes with SubAggregates in the SRCS and LBCS.

Hornfels is the dominant raw material used for lithic production in both LBCS (93.7%) and SRCS (85.7%) contexts, with chert, quartz, dolerite, and sandstone of minor importance. The relatively high frequency of retouched implements in both the LBCS (13.2%) and SRCS (12.4%) are made on hornfels. Cortex on the LBCS flakes is more angular (66.7%, n = 34) than rounded (33.3%, n = 17) compared to the SRCS where rounded cortex predominates (70.4%, n = 133) over angular cortex (29.6%, n = 56). Cortex retention is also greater on SRCS cores and flakes. Cores themselves comprise only a minor component of each assemblage; the LBCS member contains a single quartz bipolar core (Fig. 8a). Most of the SRCS cores (78%) are platform cores made on flat water-worn hornfels cobbles, flakes, and small (< 50 mm in maximum dimension) water-worn pebbles (Fig. 9a–d). The remaining 21% of SRCS cores was reduced using a bidirectional bipolar (hammer and anvil) strategy. Notably, one SRCS core was broken along the lateral edge and then flaked longitudinally along the broken edge to create elongated flakes and to refresh the edge for tasks that left the tool's edge marginally modified (see Fig. 8d). Additionally, the single LBCS core shows signs of platform-on-anvil technique (i.e., a freehand platform area opposed by a crushed distal core margin) specifically to produce elongated flakes (see Fig. 8a).

Figure 8. (color online) Sample of stone artifacts from LBCS. (a) anvil-supported bladelet core; (b) bipolar split flake with wear on the split surface; (c–e) notched flakes; (f) blade with bending fracture denoted by arrow. (a) made on quartz and (b–f) made on hornfels.

Figure 9. (color online) Sample of stone artifacts from the Shell-Rich Clayey Sands. (a) bipolar core on water-worn pebble; (b) platform core on thin water-worn pebble; (c) bipolar core on nodule; (d) core on flake with retouched edge; (e–g) naturally backed knifes; (h) large flake on cobble; (i, j) small retouched water-worn cobbles. All lithics were made on hornfels.

Flakes make up the majority of the lithic assemblages for both LBCS and SRCS. Comparisons of the SRCS and LBCS flake size (length and width) and elongation (length/width) show significantly larger and less elongated (i.e., wide and side-struck) flakes in the SRCS and smaller, more elongated flakes in the LBCS (F[1] = 15.22, p < 0.01; omega squared = 0.08). Several flakes in both the SRCS and LBCS assemblages show signs of bipolar damage (sheared bulbs of percussion and severe crushing) in their platform regions. The retouched tool frequencies are relatively high and similar between the LBCS and SRCS members. Scrapers and notched tools make up the majority of the formal tool inventory in both StratAggs. Scrapers are predominantly side and end forms, although this pattern is most common in the SRCS member with ~28% side, 21% end, and 8% double (side and end) scrapers (see Fig. 9e–g). Scraper retouch is also steeper and more invasive in the SRCS member compared to the LBCS member, where edges show shallower scarring. Only one backed tool was found in the LBCS assemblage. Additionally, at least one of the LBCS flakes shows clear evidence of a step-terminating bending fracture at its distal end suggesting use in tasks with a high degree of bending tension (see Fig. 8f).

Worked Bone

Bone is overall well preserved, but it varies in both LBCS and SRCS deposits based on proximity to the dripline where preservation is lower and sediments are largely de-calcified. Inside the dripline, bone and shell preservation rapidly improves and includes materials not limited to bone or teeth. For example, three preserved feathers have been recovered from SRCS Jess and SRCS Bonile.

A number of worked bone implements were shaped via cutting and abrasion (Fig. 10). Of particular interest is CN20667 (Lot 382) from SRCS Rosa (see Fig 10a). The implement is 25 mm long, but it is broken. One side has been abraded and cut to a conical point. Opposite the point at the thickest part of the tool are two parallel cuts that created an incision and stepped profile. The size and shape of the bone fragment is reminiscent of biconical fish gorges found in Holocene deposits elsewhere on South Africa's south coast (Bradfield, Reference Bradfield2019). Association of this piece with fishing technology, however, remains tentative.

Figure 10. (color online) Sample of modified bone implements. (a) CN20667 (Lot 382; Shell-Rich Clayey Sands [SRCS] Rosa), possibly fish gorge; (b) CN25758 (Lot 336; Light Brown Coarse Sands [LBCS] Kate) broken and facetted distal humerus of a rock hyrax (Procavia capensis); (c) CN7500 (Lot 228; SRCS Jess) bifacial bone point; (d) CN7499 (Lot 228; SRCS Jess) abraded bone point.

CN25758 (Lot 336) is from LBCS Kate (see Fig. 10b). It is a distal humerus of a rock hyrax (Procavia capensis) (S. Badenhorst, personal communication) that has been broken and facetted to create a bipointed keel on the shaft. Additionally, numerous beads have been found in SRCS Bonile and SRCS Josh that include 32 perforated tick shells (Nassarius kraussianus), perforated brown mussel shells (Perna perna), and one ostrich eggshell bead (Struthio camelus; CN26816, Lot 117; Fig. 11).

Figure 11. (color online) Perforated shells and beads from the Shell-Rich Clayey Sands (SRCS). (a) ostrich eggshell bead (Struthio camelus; CN26816, Lot 117, SRCS Josh); (b) perforated shell (Perna perna; CN26817; Lot 116, SRCS Bonile); (c) perforated shell (Perna perna; CN26818; Lot 116, SRCS Bonile); (d) perforated tick shells (Nassarius kraussianus; Lot 116, SRCS Bonile). The scale is the same for all objects in the figure.

Archaeomalacology

Archaeomalacological preservation

LBCS shells (all analyzed SubAggs) show considerable fragmentation as evidenced by a high frequency of mussel fragments lacking hinges and having flat surfaces (Fig. 12). The majority of the shell currently recovered is from the 10-mm-sieve fraction. The deepest LBCS deposits are currently located at the dripline where there is visible evidence of decalcification. In one instance shellfish has been identified at a microscopic level using SEM and XRD. The sample, which was from LBCS Kuka, contained fragmented pieces of shell with clear nacre tablets, and XRD results showed high levels of calcite and aragonite, indicating recrystallized mussel shell (see Fig. 6).

Figure 12. (color online) Examples of Light Brown Coarse Sands (LBCS) shell remains. Assorted mussel and limpet shell remains from LBCS. (A) LBCS Ollie, Lot 320; (B) LBCS Kate, Lot 330; (C) LBCS Colton, Lot 342.

Quantitatively, by dividing the total weight of each stratigraphic unit by the number of specimens (fragments and the few whole shells), SRCS Bonile is relatively less fragmented (3.7 g/specimen), followed by West (2.6), Courtney and Rosa (2.1, each), Jess upper (1.5), and Jess lower (1.4). Despite being the least fragmented among the studied SubAggs, shells from SRCS Bonile appear noticeably more burned than shells from any other SRCS SubAgg studied in this analysis, with the exception of SRCS Lwando (which was not studied). Quantitatively, 37.5% of the shell fragments from SRCS Bonile are burnt compared to <14.5% of shell fragments from the rest of the SRCS stratigraphic units. The label “burnt” refers to any visible exposure to heat, which is revealed in the change of color and/or luster of shells. For example, sometimes a shell outer surface may not become discolored but will lose its shine or gloss, while the inner surface can sometimes become beige or slightly grey in color. The green gloss that is so characteristic of brown mussels is also often lost or lessened as a result of heat exposure.

Another unique quality of the SRCS Bonile shells is that they appear to be preserving macrobotanical remains. During excavations, numerous whole and fragmentary leaves and grass stems were found adhering to the insides of mussel shells. In these cases, shells were often oriented with the cup of the shell facing downward, and so we currently believe that the shells not only shielded the macrobotanical remains but may be creating unique microenvironments that inhibit deterioration of the plant tissues (e.g., Jerardino and Navarro, Reference Jerardino and Navarro2008).

SRCS Courtney specimens are powdery (chalky) and crumbly to different degrees yet appear unburnt. It is possible that the partial loss of the shell mineralogical structure in Courtney obscures the fact that they could have also been exposed to heat as in stratum Bonile. Additionally, marine shell remains from SRCS Jess were often chalky due to decomposition of the mineralogical shell structure and brittle compared to shell found in other SRCS SubAggs. The highly fragmented nature of the shell in SRCS Jess is also evidenced by a high frequency of mussel fragments lacking hinges and showing the greatest relative levels of fragmentation. Many fragments also had flat surfaces. This pattern is consistent with higher fragmentation rates due to longer exposure of shells on the site's surface and/or compression of the deposits (Jerardino, Reference Jerardino, Giovas and LeFebvre2018).

Archaeomalacological observations

Sample sizes and percentages of marine shell species in terms of MNI and weights for SRCS samples are shown in Table 6. Sample sizes are rather small for meaningful calculations of percentages, with only two stratigraphic units in SRCS represented by numbers above 100 individuals, namely SRCS West (MNI = 104) and Bonile (MNI = 267). The LBCS samples await full analysis, but some preliminary observations are available. The total shell fragment counts (NISP) are as follows: LBCS Chopi (21), LBCS Hayley (19), LBCS Kuka (shell remains only identified via SEM and XRD), LBCS Ollie (53), LBCS Kate (44), and LBCS Colton (8). No additional archaeomalacological observations are currently available for LBCS samples from the 2019 excavated deposits that greatly expanded the sample of LBCS materials.

Table 6. SRCS Shellfish. This table shows the species and common names of shellfish identified in Shell-Rich Clayey Sands (SRCS) deposits. The species are subdivided to show relative minimum number of individuals (MNI) and percentages by SubAggregate.

Among the studied SRCS samples, 17 shellfish taxa were identified to genus and to broad taxonomic groups (see Table 6). Brown mussel (Perna perna) is the dominant species in the assemblage (70.0%–86.5% by MNI), with minor but noticeable contributions of small barnacles, oysters, and limpets. Taking into account the small number of measurable individuals (total n = 90, including left and right valves), most of the collected brown mussels that were brought back to Waterfall Bluff appear to be rather small or medium in size, measuring between 40 mm and 80 mm. Initial observations from LBCS samples appear to reflect the same species abundance profile, but full analyses will test these first observations and provide quantified data to compare with the younger SRCS samples. Other much-less-frequent species in the Waterfall Bluff record, like Comus platyaulax, Nerita albicilla, and Mitra litterata, may have been picked up by chance during foraging activities or as potential personal ornaments. Others are known to arrive to campsites clinging to other shells (i.e., Helcion sp., keyhole limpets, and barnacles) or attached to the byssus or threads of rocky shore mussels (i.e., Barbatia obliquata) (Jerardino, Reference Jerardino2017). The latter species are all generally too small (<40 mm) for food consumption.

Archaeoichthyofauna

Temperature, geology, and biological interactions are the three main influences on South African fish distributions, but temperature is considered the most significant (Penrith, Reference Penrith1970; Turpie et al., Reference Turpie, Beckley and Katua2000). There is an east–west gradient in fish species richness due largely to the impact of the warmer-water Agulhas current on the east coast, which provides a high-latitude habitat for tropical Indo-Pacific fish species within South Africa's subtropical east-coast waters (Turpie et al., Reference Turpie, Beckley and Katua2000; Maggs, Reference Maggs2011). The east coast also has high species richness due to increased diversity in inshore habitats (Smith, Reference Smith, Bruton and Cooper1980). Pondoland's sub-tidal habitat is composed of coast-parallel shelving reef complexes. There is a shift in benthic composition from algal-dominated reefs in the north to sponge- and ascidian-dominated communities in the south, with a major shift from Mbotyi southwards (Celliers et al., Reference Celliers, Mann, Macdonald and Schleyer2007). Celliers and colleagues (Reference Celliers, Mann, Macdonald and Schleyer2007) attribute this shift to higher nutrient load from riverine runoff in the south.

Today, eight species of fish account for 90% of the total estuarine fish catch (Maggs et al., Reference Maggs, Mann and Cowley2013): Chrysoblephus puniceus (slinger seabream; Sparidae), Galeichthys trowi (Natal sea catfish; Ariidae), Polysteganus praeorbitalis (Scotsman seabream; Sparidae), Lethrinus nebulosus (green snapper; Lethrinidae), Epinephelus marginatus (dusky grouper; Serranidae), Epinephelus andersoni (cat-face rock-cod; Serranidae), Epinephelus rivulatus (halfmoon grouper; Serranidae), and Cymatoceps nasutus (black musselcracker; Sparidae). These fish belong to two main families: Sparidae and Serranidae.

Archaeoichthyofaunal methods

The remains described here include plotted bones recovered from the 10 mm and 3 mm sieves. Fish remains from the 1.5-mm sieve have not yet been studied. Skeletal elements terminology and calculations (NISP, %) were performed following standard zooarchaeological approaches and methods (Wheeler and Jones, Reference Wheeler and Jones1976; Grayson, Reference Grayson1984; Cannon, Reference Cannon1987; Rojo, Reference Rojo1991; Reitz and Wing, Reference Reitz and Wing1999; Ramsey, Reference Ramsey2009b). To visualize presentation of cranial and postcranial regions we grouped the skeletal elements according to these two groups and by eight anatomical regions (Butler, Reference Butler1992; Amante and Eakins, Reference Amante and Eakins2009).

NISP was used as a basic quantitative unit for measuring counts, relative abundance, ordinal ranking, body part representation, spatial distribution, and taxonomic composition. To aid our taxonomic identifications, we have partnered with the East London Museum, the South African Institute for Aquatic Biodiversity, the Eastern Cape Parks and Tourism Agency, and the East London Aquarium to develop a reference collection of local fishes. Since the collection has only just started, we have not been able to identify most remains. We were able to do preliminary taxonomic identification for a few diagnostic bones (dentary, premaxilla) that were indicative at the family level.

Measurements were performed with a digital caliper on vertebrae centrum height, width, and length (Morales Muñiz and Rosenlund, Reference Morales Muñiz and Rosenlund1979; Zohar et al., Reference Zohar, Dayan and Spanier1997; Zohar and Cooke, Reference Zohar and Cooke2019), and on dentary and premaxilla length and height (Fig. 13). We also recorded the color of the bone, and signs of burning were identified on charred bones displaying black, gray, and white colors.

Figure 13. (color online) Early and Middle Holocene fish vertebrae dimensions. This figure shows the distribution of fish vertebrae centrum width for Early Holocene and Middle Holocene assemblages with reference photos showing how the measurements were collected. The data suggest larger fish were collected during the Early Holocene.

Archaeoichthyofaunal results

Over 1,600 fish remains (NISP) have been identified to date from Waterfall Bluff, spanning LBCS SubAggs Kuka, Chopi, and Lily as well as all SubAggs in the SRCS. Most of the marine fish remains (69%; NISP = 1,108) are from SRCS SubAggs dated to the Early Holocene. The next greatest concentration of fish remains are found in the SRCS SubAggs dated to the Middle Holocene (31%; NISP = 492). Marine fish remains are also present in the LBCS (NISP = 4; 0.02%), but the sample is currently very small due to the limited LBCS excavations and sample that was available at the time of the analysis; the sample has been expanded since then.

Analysis of 767 skeletal elements sampled from SRCS show the presence of 73 different skeletal elements from the cranial and postcranial regions, including a single otolith from a very large fish (not yet identified) (Supplementary File 6). These data show that although the most abundant skeletal elements are vertebrae (37%, NISP = 287), in most SubAggs the cranial region is not underrepresented (Fig. 14). The poor preservation of bones from the branchial region is not surprising and supports our assumption that the fish were gutted in a different location, probably on the beach, close to their area of capture (Morales Muñiz and Rosenlund, Reference Morales Muñiz and Rosenlund1979; Zohar and Cooke, Reference Zohar and Cooke1997, Reference Zohar and Cooke2019). Additionally, many of the fish remains are well preserved, including preservation of fragile features such as the vertebrae with haemal and neural spines (see Fig. 13 and Fig. 15).

Figure 14. (color online) Percentage of preserved fish crania vs. postcrania. This chart shows the relative percentage of crania vs. post-crania in the Waterfall Bluff Shell-Rich Clayey Sands (SRCS) fish assemblage from 9 SubAggregates dating to the Middle Holocene and Early Holocene.

Figure 15. (color online) Examples of preserved fish bones from Waterfall Bluff. (a) Fish spine (CN471, Lot 107 Shell-Rich Clayey Sands [SRCS] Nate); (b) Sparidae premaxilla (CN10932, Lot 106, SRCS Nate); (c) vertebra CCN12124; Lot 115, SRCS Josh); (d) occipital crest (CN26765, Lot 266, SRCS West); (e) vertebra (CN2167, Lot 107, SRCS Nate); (f) hyomandibular (CN493, Lot 114, SRCS Bonile); (g) 1^st dorsal pterygiophore (CN466, Lot 123, SRCS Josh).

Species-specific measurements of fish vertebra are well known to correlate to the overall size of the fish (Granadeiro and Silva, Reference Granadeiro and Silva2000). We measured the vertebrae centrum diameter (maximum length, width, and height) on a small sample (NISP = 133) from SRCS SubAggs attributed to either the Early Holocene (NISP = 88) or Middle Holocene (NISP = 45) (see Fig. 13). This analysis combines different species, so it only provides preliminary information on the relative sizes of the fishes that are represented in the assemblage. The observed SRCS pattern shows that all size groups of fishes are present, ranging from small (less than 10 cm in total length) to large fish (> 70 cm in total length). Larger fish are more abundant in the Early Holocene sample compared to the Middle Holocene sample (see Fig. 13). Future analysis will enable us to examine whether the reduction in fish size observed between Early and Middle Holocene samples accords with changes in fish population or changes in human exploitation patterns, or if it is biased due to sample sizes.

Taxonomic identification is currently in progress. So far, identifications at the family level include the following (in descending order): Sparidae (sea breams and porgies), Mugilidae (mullets), Sciaenidae (drums), Haemulidae (grunters), Serranidae (groupers), Carangidae (jacks), and Labridae (wrasses). Species of these families are among the most common fish captured in this area today (Maggs, Reference Maggs2011), and they are particularly common in littoral zones within estuaries and can tolerate both brackish and salt water. Additionally, we have identified the remains of crab (spp. unidentified) and spiny lobster (spp. unidentified).